|
VPAR FAQ
GENERAL
- ¿which countries does VPAR recognize.
Check the updated country list here .
- ¿what does NeuralLabs need to "train" a new country?
We need 500 sample images. These images have to be taken with a good quality and good resolution in order to make a good training.
(See "Which requirements must the images have in order to be processed properly by VPAR?" below)
- Does VPAR read ALL the countries?
Our philosophy to get spectacular results is to localize the engine version for each country.
So we have separate versions depending on the country. These versions can also read number plates from surrounding countries, but it will get almost perfect results with the plates it is trained to read.
-¿Does VPAR recognize trucks number plates?
Yes. In addition, if you activate the "trucks parameter" on initialization, the engine will return both number plates (truck and tow) if you are reading from behind.
-¿Does VPAR recognize 2 lines number plates?
YES.
IMAGE CAPTURE
-Does VPAR recognize moving cars?
YES. The speed of the vehicle has to be with the kind of camera, lighting, frame grabber, etc. but not with our engine.
You will have to get a "good image" to send it to our engine. So the engine is hardware independent.
The speed of the vehicle will tell you which hardware to use.
As an example, in high speed situations our customers use progressive scan cameras. (Police cars, highways, etc)
On the other hand, in parking systems where the vehicle usually stops to get a ticket, our customers use normal interlaced cameras.
-Which cameras does VPAR support?
VPAR is HARDWARE independent. Because it's a just an engine it only needs one image.
The origin of this image is not important.
The cameras can be interlaced, progressive scan, analogical, IP, etc.
The engine will process bmp, jpg files or RGB8, RGB24, RGB32 bits buffer directly from the camera.
- Which requirements must the images have in order to be processed properly by VPAR?
- Images should be as "clear" as possible.
- VPAR requires a MINIMUN resolution. Characters of the number plate must have 25 pixels height. We recommend 30 or more but 25 is enough. If characters have less that 25 pixels VPAR will read then anyway but the reliability will be lower.
- That means that image size is not important. The only important think is the character height.
- Should be avoided, as far as possible, number plate character occlusions because of objects between the camera and the vehicle, strong shadows on the number plate, shine on the number plate.
- The images should be taken with the camera as much as possible perpendicular to the plane of the number plate to minimize the effects of distortion in perspective, tilt, etc
DEVELOPMENT
-Which programming languages can be used with VPAR?
Any programming language that can use a windows DLL (dynamic link library).
Visual Basic, C++, c# , .net, Borland c++ builder, FoxPro, Delphi, etc.
-Does VPAR recognize JPG Files?
Yes. But we must remember that the JPEG format is a format with missing information. This means that it sacrifices image quality (detail) to achieve a higher compression ratio (smaller file). We should try always to use images without JPEG compression. If it is not possible we should try to lower the level of compression to the minimum (i.e. the highest possible quality).
- Why sometimes the processing time is quite high?
VPAR can automatically detect the height of the characters in the number plate. However if you use this parameter this will make the recognition slower.
Our advice is:
In phase of testing, use automatic detection. (-1)
Using the function vparespGetAverageCharacterHeight you will get the heights that the system is detecting. (We can make a log to store them)
These heights can be used to calculate the average height of the characters for that camera (in a concrete situation).
Use this average value to set up character height on the engine (instead of automatic height detection).
Thus the system will work much faster.
LICENSING
-Can I recognize images of several cameras with a single license?
Yes. However, we do not recommend using more than 4 cameras with a single computer (with one license) in order to give an acceptable response time. The appropriate number of cameras that you can process with one computer will depend on the hardware used, the parameters of recognition, quality the images of traffic density, etc....
Which is the licensing policy?
One license per computer.
SPECIAL VERSIONS
- What if I need more processing speed?
Neural labs has published in December 2007 its HS (high speed) version for environments where the speed of processing is an essential requirement.
This version is up to 55% faster than the Standard version.
- VPAR PLATFORMS
Vpar can be provided for platforms such as Linux, QNX, etc..
This is because the code is entirely in C + + Standard. Thus it is not dependent on any platform and is easily migrable any other operating system as necessary.
|
